arrow
Return

Improving large models with small models: Lower costs and better performance

delete2025-11-04
delete0
PRE
AI
陈董 cover
陈董 (Dong Chen)
F
Fei Gao
S
Shuo Zhang
庄越挺 (Yueting Zhuang)
汤斯亮 (Siliang Tang)
Q
Qidong Liu
H
Hua Wang
杨鑫 (Xin Yang)
M
Mingliang Xu
DOI:10.1016/j.neunet.2025.108276delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
• A new paradigm improves large model performance and reduces costs by introducing specialized small models. • Introduce S4L and L4S to further enhance large model performance while reducing costs. • Compared with fine-tuning, the proposed method offers a new perspective on injecting specific knowledge. • Demonstrates the generality through integration with both open-source and closedsource large models.

Journal

Neural Networks cover
Neural Networks
IF:
6.3
Papers:
7.8K
Citations:
3.0W

Organization

Z
Zhengzhou University
Scholars:
6.8W
Papers: 4.4W
Citations: 8.5W
Z
zhejiang university
Scholars:
17.5W
Papers: 12.0W
Citations: 152
researcher View more organizations